Hindawi Publishing Corporation

EURASIP Journal on Applied Signal Processing

Volume 2006, Article ID 18083, Pages 1–20
DOI 10.1155/ASP/2006/18083

A MIMO-OFDM Testbed for Wireless Local Area Networks

Albert Guillén i Fàbregas,1 Maxime Guillaud,2 Dirk T. M. Slock,2 Giuseppe Caire,3 Karine Gosse,4
Stéphanie Rouquette,4 Alexandre Ribeiro Dias,4 Philippe Bernardin,4 Xavier Miet,4 Jean-Marc Conrat,5
Yann Toutain,6 Alain Peden,7 and Zaiqing Li7
1 Institute for Telecommunications Research, University of South Australia, SPRI Building, Mawson Lakes Boulevard,
Mawson Lakes, SA 5095, Australia
2 Eurécom Institute, 2229 Route des Crêtes, 06904 Sophia Antipolis, France
3 Electrical Engineering Department, University of Southern California, 3740 McClintock Avenue, Los Angeles, CA 90089, USA
4 Motorola Labs, Centre de Recherche de Motorola, Espace Technologique Saint-Aubin, 91193 Gif-sur-Yvette, France
5 France Telecom R&D, 3 avenue des Usines, 90007 Belfort, France
6 Antennessa, avenue la Perouse, 29280 Plouzane, France
7 École Nationale Supérieure des Télécommunications de Bretagne, Technopôle Brest-Iroise, 29238 Brest, France

Received 15 December 2004; Revised 11 May 2005; Accepted 21 June 2005

We describe the design steps and final implementation of a MIMO OFDM prototype platform developed to enhance the perfor-
mance of wireless LAN standards such as HiperLAN/2 and 802.11, using multiple transmit and multiple receive antennas. We first
describe the channel measurement campaign used to characterize the indoor operational propagation environment, and analyze
the influence of the channel on code design through a ray-tracing channel simulator. We also comment on some antenna and RF
issues which are of importance for the final realization of the testbed. Multiple coding, decoding, and channel estimation strategies
are discussed and their respective performance-complexity trade-offs are evaluated over the realistic channel obtained from the
propagation studies. Finally, we present the design methodology, including cross-validation of the Matlab, C++, and VHDL com-
ponents, and the final demonstrator architecture. We highlight the increased measured performance of the MIMO testbed over
the single-antenna system.

Copyright © 2006 Albert Guillén i Fàbregas et al. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

1. INTRODUCTION networks (WLAN) is then a promising environment for the

Future wireless communication networks will need to sup-
port extremely high data rates in order to meet the rapidly
growing demand for broadband applications such as high-
quality audio and video. Existing wireless communication
technologies cannot efficiently support broadband data rates,
due to their sensitivity to fading. Multiple antennas have
recently emerged as a key technology in wireless commu-
nication systems for increasing both data rates and system
performance. Ever since the landmark results by Telatar [1]
and Foschini and Gans [2], there has been a growing inter-
est in developing practical transmission schemes that exploit
the extra spatial dimension to achieve reliable communica-
tion over multiple-input multiple-output (MIMO) channels.
In particular, the application of such techniques to current
wireless communication systems is of great interest. High-
throughput communication in indoor wireless local area
deployment of multiantenna equipments.
The main scope of this paper is to present the design steps
leading to the demonstration of a software-radio testbed with
multiple transmit (Tx) and receive (Rx) antennas, aiming
at enhancing the performance of current WLAN standards.
The expected benefits of multiple antennas are subject to the
propagation conditions in wireless indoor local area trans-
missions. Therefore, the propagation environment must be
carefully studied and analyzed before appropriate multiple-
antenna transmission techniques are devised and evaluated.
In this regard, the scope of our work is larger than the typical
prototyping efforts in that the project resulted in the produc-
tion of an entirely custom demonstrator by the various in-
dustrial partners, including the algorithmic studies, the hard-
ware handling the baseband processing, the RF components,
up to the antennas, and the working real-time implementa-
tion of the coding and decoding algorithms. In particular,
2 EURASIP Journal on Applied Signal Processing
the channel characterization effort and the corresponding al-
gorithmic adaptations proposed in this work, are typically
absent from smaller-scale prototyping projects. This charac-
terization was done by measuring the channel parameters in
the frequency band of interest and developing channel mod-
els as well as corresponding channel simulation capabilities
in order to produce typical channel impulse responses for
algorithms evaluation. Then, using this preliminary infor-
mation on propagation characteristics, various multiantenna
processing techniques have been studied and compared ac-
cording to performance versus complexity criteria. Similarly,
the design and use of custom RF components and anten-
nas are often replaced by channel simulators. Another major
point in the validation of the testbed performance concerns
the careful comparisons and cross-validations between hard-
ware testbed performance and software simulation results. In
the sequel, we give some details on the aforementioned steps
and illustrate the design of the testbed as well as its perfor-
mance improvement with respect to the single-antenna case.
The paper is organized as follows. Section 2 addresses
the current status of WLAN specifications and the position-
ing of the paper in that respect. Section 2.1 presents the
modeling of the propagation environment, and Section 2.2
shows the analysis of the propagation medium and its impact
MIMO system design. Section 3 highlights some considera-
tions about antennas and RF, in the view of future integration
and production. Section 4 describes the several candidate al-
gorithms for implementation on the testbed. In particular,
multiple-antenna processing techniques in both open-loop
and closed-loop modes and channel estimation algorithms
are described. Section 5 provides details of the demonstra-
tion testbed as well as the performance improvement that
was obtained thanks to the exploitation of multiple spatial
streams. Finally, Section 6 presents the conclusions of this
work.
2. MIMO WLAN ENVIRONMENT
After the success of WLAN products based on IEEE802.11b,
the higher-rate WLAN solutions based on the IEEE802.11a/g
standards have started to spread on the market. Such WLAN
systems are based on orthogonal frequency-division mul-
tiplexing (OFDM), whereby the transmitted signal is con-
structed as the inverse Fourier transform of the actual con-
stellation symbols to transmit. OFDM turns a frequency-
selective channel into parallel flat-fading subchannels, each
corresponding to a frequency subband of the total signal,
with the benefit that simple equalization methods can be
applied. Higher-order modulations, from BPSK up to (op-
tional) 64-QAM enable to reach bit rates up to 54 Mbps, in a
20 MHz band. These WLANs operate in the 2.4 GHz indus-
trial, scientific, and medical (ISM) band for the .11g OFDM
solution and in the 5 GHz unlicensed band for the .11a air
interface. In this paper, we focus on the 5 GHz band that rep-
resents the majority of the available spectrum.
Similar standardization efforts on WLAN systems have
taken place in Europe and Japan simultaneously, giving rise
to the ETSI HiperLAN/2 and ARIB MMAC specifications,
that rely on exactly the same transmission principles as
IEEE802.11a for the physical layer. However, because of the
absence of existing products following the European specifi-
cations today, this paper refers to the IEEE802.11a context.
In this work, some initial choices for the testbed were made
based on HiperLAN/2-specific characteristics, such as the
preambles used for channel estimation and synchronization.
However, this does not affect the generality of the approach
and the final performance results.
IEEE802.11a/g systems were initially developed taking
into account a single-antenna constraint. Once the interest
for MIMO systems gained in importance, multiple-antenna
products appeared, which did not make full use of the
multiple-antennas framework as envisioned by Telatar or
Foschini. Essentially, they had to remain single-stream sys-
tems, that is, transmitting only one symbol per channel use,
in order to remain compliant with the standard. In this case,
the transmission does not benefit directly from the increased
channel capacity promised by multiple antennas, but merely
exploits the diversity in order to increase the link quality.
Methods such as selection diversity or beamforming at the
transmitter, and various types of combining or interference
cancellation at the receiver were used.
Finally, with the clear recognition of the advantage of us-
ing multiple transmit and receiving antennas in terms of per-
formance (link quality, range, bit rate), a new task group was
created in 2003 within IEEE802.11, as .11n, with the char-
ter of designing new WLAN specifications achieving up to
100 Mbps on top of the media access control (MAC) layer in
20 MHz band at 2.4 GHz but mainly at 5 GHz. There, the key
new technology introduced to enhance .11a performance is
the MIMO signal processing capability. The other system pa-
rameters discussed today are very similar to the IEEE802.11a
ones (64-point IFFT OFDM modulator for 20 MHz band,
baseline convolutional encoder). As a consequence, the work
presented in this paper is perfectly relevant for the ongoing
progress of standardization in the area of next-generation
MIMO WLANs.
2.1. Propagation environment modeling
and simulation
In order to evaluate the performance of radio access tech-
niques, realistic channel modeling is required. To this end,
a measurement campaign was performed in both outdoor
(campus) and indoor operational environments at 5 GHz, in
order to build a wideband 3D model of the channel (only the
indoor measurements are presented here, further informa-
tion about this campaign can be found in [3]). These mea-
surements were further analyzed, in order to build a software
channel simulator that can accurately replicate the environ-
ment of an indoor MIMO WLAN.
The MIMO channel was measured using two antenna ar-
rays. A virtual array is formed by moving the Tx antenna
(Figure 1(a)) on a grid composed of two horizontal and or-
thogonal 17-sensor linear antenna arrays (Figure 2(a)). The
Tx antenna is omnidirectional and was set at 1.2 m above the
floor corresponding to a mobile terminal (MT) location. The
Albert Guillén i Fàbregas et al.
(a)
Figure 1: On-location channel sounding equipment: (a) mobile terminal and (b) access point.
λ/4
4λ
Tx horizontal
cross-linear
antenna array
(a)
Figure 2: (a) Tx and (b) Rx antenna arrays configuration.
Rx antenna array (pictured in Figure 1(b)) is an 9 × 9 vertical
planar antenna array (Figure 2(b)). The horizontal and verti-
cal apertures of the Rx antenna are, respectively, equal to 120◦
and 90◦ . An additional fixed omnidirectional antenna is used
to check the stationarity of the propagation channel during
a full MIMO measurement. The Rx antenna was set at 2.6 m
above the floor corresponding to an access point position.
For each MT position, a full MIMO measurement consists of
34×81 complex channel impulse responses (CIR). CIRs were
measured at 5.2 GHz with a temporal resolution equal to
4 nanoseconds. The transmitted and received polarizations
were vertical. The wideband measurements were performed
by the Americc channel sounder developed by France Télé-
com R&D.
Two Rx locations were chosen as representative of WLAN
access point (AP) position. The first one was located in a cor-
ridor surrounded with rooms separated by very thin walls or
glass doors. The second Rx position was located in a corridor
surrounded with rooms separated by much thicker walls and
3
(b)
3.2 λ
2λ
Rx vertical planar
antenna array
(b)
wood doors. Twenty four transmitter locations were cho-
sen as representative of MT positions (office, large meet-
ing room, corridor, etc.). The maximum Rx-Tx distance was
45 m.
The data analysis phase consists of a space-time data
processing applied to the measurements corresponding to
each Tx-Rx location. The power angular profiles at the AP
and MT were computed for each excess delay, using a stan-
dard beamforming algorithm. An example of 3D space-time
diagrams at the AP is given in Figure 3. We can observe
that the power delay profile (PDP) has the classic exponen-
tially decaying shape often typical of indoor radio chan-
nels. Synthetic wideband and spatial parameters were also
computed. For instance, the mean delay spread, AP azimuth
spread, and AP elevation spread on the whole measurement
campaign are, respectively, 19.2 nanoseconds, 20.2◦ , 14.4◦ .
The radio channel angular and frequency-selectivity analysis
shows that the ETSI BRAN [4] models disagree with the re-
sults from the operational environments characterized in this
4 EURASIP Journal on Applied Signal Processing

Azimuth (degree)

Amplitude (dB)
0
−10
−20

−30

60 60
40 40
Elevation (degree)

Elevation (degree)
20 20
0 0
−20 −20
−40 −40
−60 −60
−30 −20 −10 0 0 50 100 150 200 250
250 250 Delay (ns)
0
200 200
Delay (ns)

Delay (ns)
150 150 −10
100 100

50 50 −20

0 0
−30 −20 −10 0 −60 −40 −20 0 20 40 60 −30
Amplitude (dB) Azimuth (degree)

Figure 3: Space-time diagrams for a typical NLOS situation, 0◦ corresponds to the antenna array boresight, delay spread, AP azimuth spread,
AP elevation spread.

experimentation. In office environment, delay spread (DS)
values are significantly smaller than the values recommended
by Model A. The values of DS and AP azimuth spread (AS)
strongly depend on the radio link configuration, which can
be either line-of-sight (LOS) or non-line-of-sight (NLOS).
No significant correlation was found between these two syn-
thetic parameters and other parameters like Tx-Rx distance
or MT position (corridor, office, laboratory, etc.). Two wide-
band geometry-based propagation channel models are pro-
posed for MIMO transmission simulations in indoor en-
vironment at 5.2 GHz: one for LOS situations and an an-
other one for NLOS situations (Figure 1(b)). These models
are wide sense stationary uncorrelated scattering (WSSUS)
models and reproduce a selected MIMO measurement con-
sidered as typical. The developed models reproduce accu-
rately the temporal dispersion and the angular dispersion
seen at the AP. At the MT, the angular dispersion was only
investigated by space-time diagrams visual inspection and a
uniform 3D distribution was chosen to model the direction
of arrival (DOA) at the MT. The MT DOA model could be
refined by more accurate data analysis. Further information
on geometrical models, and a selection of typical cases can
be found in [5, 6].
The propagation channel models have been implemented
using the Mascaraa ray-based radio channel software sim-
ulator [7], which simulates the polarized space-time be-
havior of the propagation channel. Therefore, the propa-
gation channel is represented by a large set of rays with
geometrical and propagation characteristics. Geometrical
characteristics are delays and directions of arrival at the re-
ceiver and transmitter. The propagation characteristics are
attenuations computed as functions of the transmitted and
received polarization. The ray-based (so-called physical) ap-
proach has various advantages among which the most rele-
vant are as follows:
(i) a set of rays includes all the information on the channel
and every usually defined mathematical variable that
can be obtained from the set of rays (channel impulse
responses, Doppler, angular spectrum, correlation ma-
trices, etc.),
(ii) most of the wideband channel models can be simu-
lated with this simulation tool (tapped delay line mod-
els, tap directional models [8], scatterer or geometrical
models, ray-tracing or ray-launching models [9]),
(iii) a set of rays models the propagation channel regardless
of the antenna gains.
The channel simulator provides a large set of functionalities
to facilitate the propagation models integration in simulation
chains. The most important are the rays generation from a
given channel model; the impulse response processing tak-
ing into account the 3D vectorial antenna pattern of each
sensor; the convolution of the input signal with the simu-
lated impulse response. The channel simulator is designed in
a modular way and can be incorporated to link level simula-
tion tools for radio-communication systems such as UMTS,
HIPERLAN, WiFi, and so forth, with or without multiple
antennas.
Albert Guillén i Fàbregas et al.
Frequency domain channel average autocorrelation
1 channel model has much larger frequency diversity, and as
0.8
0.6
0.4
0.2
0 a correlation of 0.14. At the AP, the correlation figures differ
−10 −5 0 5
Frequency (MHz)
BRAN-A model
Simulated indoor channel
Figure 4: Correlation versus frequency spacing for the BRAN-A
and the simulated WLAN channels.
2.2. Impact of propagation characteristics on
MIMO system design
In order to extract design criteria for space-time and space-
frequency code design, simplifying assumptions on the
channel statistics have to be made. In particular, a common
assumption is that the channel can be described by a finite
number of i.i.d., Gaussian random variables, this number
being referred to as the channel diversity. Most codes are built
to exploit the maximum diversity of the channel model. In
order to gain a better understanding of the properties of a
typical WLAN channel, an extensive measurement campaign
was made, and a channel simulation tool was developed, as
detailed in Section 2.1. We now present an analysis of the fre-
quency and space diversity of the considered WLAN channel.
The measurement-based channel simulator described above
was used to generate channel realizations corresponding to
an indoor, NLOS propagation model typical of an office en-
vironment, with 2 Tx and 2 Rx antennas. In an effort to re-
main close to the scenario of a mobile communicating with
its AP, one side of the transmission link (the MT) has om-
nidirectional antennas, whereas the other side (the AP) has
sectorial antennas with a 120◦ aperture.
Statistical analysis of the simulated channel showed that,
despite the rich multipath promised by the NLOS configu-
ration, the indoor setting results in a small delay spread—
of the order of two or three sample periods. Indeed, it was
measured that the two strongest taps of the impulse response
account on average for more than 90% of the power carried
over the channel. The consequence is the limited frequency
diversity available from such channel, that is, the strong cor-
relation of the channel coefficients between subbands. This is
depicted in Figure 4 as a function of their spacing in the fre-
quency domain for a given transmit-receive antenna pair. It
can be observed that correlation remains very high, even for
5
frequencies situated far apart in the spectrum. The BRAN-A
stated before, it does not accurately model a typical indoor
scenario.
Another concern with MIMO systems is the antenna
spacing necessary to ensure that different antennas actually
yield uncorrelated channels. The AP-side (resp., mobile-side)
correlation is defined as the off-diagonal coefficient of the
normalized covariance matrix E[HH H
s Hs ] (resp., E[Hs Hs ]),
where Hs is the 2 × 2 MIMO channel associated with the
strongest downlink path. At the MT side of the link, an-
tenna separation of λ/2 yields a correlation of 0.58, whereas
a separation of one wavelength (λ = 6 cm at 5 GHz) yields
10 due to the different antenna diagrams. They are, respectively,
0.25, 0.24, and 0.25 for separations of λ/2, λ, and 3λ/2. The
critical value is the MT antenna spacing, since the size con-
straints at the access point are presumably less stringent. A
one-wavelength separation yields adequate channel correla-
tion. According to [10], whereby the assumption of a normal
distribution of scatterers of variance σα2 around the broadside
axis yields a correlation ρ(Δ, σα ) ≈ e(−1/2)(2πΔσα ) , these values
2
correspond to values of σα between 10 and 20◦ . This result,
◦
associated with the low delay spread mentioned above, seems
to denote a channel with few separable paths.
The ergodic capacity of a MIMO channel with NT trans-
mit and NT receive antennas with complex Gaussian i.i.d.
fading coefficients was shown by Telatar [1] to scale linearly
with min(NT , NR ) while the signal and noise powers remain
constant. This is a very important property in favour of
MIMO transmission, since not all system parameters have
such an influence on the link capacity. Increasing the trans-
mission power, for instance, only has a logarithmic influence
on the capacity. We verified the Gaussianity of the chan-
nel coefficients by means of a χ 2 analysis. Results revealed
that they can be accurately modeled as Gaussian random
variables. On the other hand, independence of the chan-
nels between different antenna pairs is by no means guar-
anteed in reality, since they essentially all depend on the
same physical environment. This is a concern since corre-
lation among channel coefficients can reduce the capacity.
In extreme cases, such as the pinhole [11] (also called key-
hole [12]) channel it was shown that the channel transfer
matrix can be rank-deficient even though its coefficients are
uncorrelated. In order to evaluate the impact of this phe-
nomenon, we compared the capacity of the channel gener-
ated by the Mascaraa channel simulator, with that of a MIMO
flat-fading channel with i.i.d. coefficients. This evaluation
was done for a signal-to-noise ratio (SNR) of 30 dB, for the
cases NR = NT = 1, 2, and 4 antennas.
Note that both, the channel statistics and the nature of
the channel are different for the i.i.d. Rayleigh and for the
simulated channels. In fact, while one is i.i.d. in space, the
other has correlation patterns at both, transmitter and re-
ceiver ends. Moreover, the i.i.d. channel is flat and the sim-
ulated channel has a nonnegligible delay spread. Figure 5(a)
shows the evolution of the ergodic capacity and the 1% out-
age capacity versus the number of antennas. The ergodic
6 EURASIP Journal on Applied Signal Processing
1000
900
800
700 Ergodic capacities
Capacity (nats)
600
500
400
300
1% outage capacity
200
100
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
NT = NR
Simulated channel
i.i.d. flat-fading channel
(a)
Figure 5: (a) Capacity evolution versus number of antennas and (b) capacity cumulative density functions.
capacity of the simulated channel follows closely the linear
curve of the ideal (i.i.d. flat-fading) channel. Interestingly,
the 1% outage capacity curves also show a linear increase
with the number of antennas, with the simulated channel
behaving better than the i.i.d. model. The capacity cumu-
lative density function (CDFs) for these cases, depicted in
Figure 5(b), provide more insight into this phenomenon: on
average, in terms of capacity, the channel behaves close to
a flat, Gaussian i.i.d. channel. A discrepancy between the
curves is clearly visible in the lower probability range, that is,
for the least likely cases, where the simulated channel is better
than the flat i.i.d. channel. A likely explanation for this phe-
nomenon is that the time-diversity of the channel, although
reduced as we have seen before, mitigates the influence of
the improbable draw of a very bad channel. A similar dis-
crepancy is also visible at the high probability range, for the
NR = NT = 4 case. In this range, the simulated channel per-
forms slightly worse than the ideal flat Gaussian i.i.d. model.
3. ANTENNA AND RF CONSIDERATIONS
The design of antenna arrays for MIMO systems can be op-
timized to maximally exploit the channel. This has been in-
vestigated in [13], where the considered optimization crite-
rion was the ergodic capacity, in order to maximize bit rate.
Such an optimization requires a doubly stochastic channel
model: it should not only model the fast fading due to the
superposition of many micropaths around a nominal path,
but it should also describe the distribution of the nominal
paths, since the multipath configuration in the actual deploy-
ment scenario is unpredictable. For these reasons, a stochas-
tic channel model is required with a uniform distribution of
angles of arrival and departure, possibly limited to a sector
(oriented away from the wall) for the case of the AP. With
1
0.9
0.8
0.7
0.6
CDF
0.5
0.4
0.3
NT = NR = 1 NT = NR = 2 NT = NR = 4
0.2
0.1
5 0 100 200 300 400 500 600 700 800 900 1000
Capacity (nats)
Simulated channel
i.i.d. flat-fading channel
(b)
such a uniform angular distribution, the pattern optimiza-
tion for an array of (colocated) antennas leads to the follow-
ing conclusions.
(i) The individual antenna responses should be orthogo-
nal (uncorrelated).
(ii) The antennas jointly should form a uniform parti-
tioning of the angular space (equal weight combining
should lead to an omnidirectional response).
(iii) The antennas should introduce a sampling of the an-
gles that is as dense and as uniform as possible. In this
way, every antenna will contribute usefully, irrespec-
tive of the multipath scenario.
The orthogonality of antennas can be obtained by two mech-
anisms: either by amplitude variation or by phase variation.
The first solution could be constructed on the basis of very
thin sectoring into N uniform sectors. Then, in order to con-
struct an array with M sensors, the diagram of one sensor
consists of the union of the sectors obtained by taking one
sector every M (consecutive) sectors—this corresponds to a
subsampling operation. In the case of phase variation, each
sensor would be a spatial all-pass filter, characterized by its
phase response. In practice though, it appears to be impos-
sible to have phase variation without amplitude variation
(similar to the minimum-phase property of linear systems).
Therefore, the sector-based solution appears more realistic.
A minimal solution would be to use one lobe or sector per
sensor. However, such a solution is not very robust since it
may be that very few paths pass in a given sector. To robus-
tify this solution, a sensor with two lobes can be considered,
centered around the same axis but oriented in the two op-
posite directions of this axis. If lobes with an opening of 60◦
are considered, then the horizontal plane can be covered with
three such sensors, as shown in Figure 6(a), with the angles
Albert Guillén i Fàbregas et al.
(a)
Figure 6: Pattern diversity with a double-lobe antenna array with 4 sensors. (a) 3 double-lobe responses and (b) slit antenna.
(a)
Figure 7: (a) Patch antennas for the MT and (b) antenna array for the AP.
between the three axes at 60◦ . A fourth such sensor can be
added with a vertical axis. This would yield a system with 4
outputs, covering all directions with a certain robustness, due
to the double sectors per sensor. An antenna whose diagram
has the two opposite lobes mentioned above can be realized
by a slit antenna, shown in Figure 6(b). The colocalization of
four of such antennas poses a problem though.
The cost and performance issues associated with the
mass-manufacturing of a multiple-antennas terminal were
investigated. For example, the cost of integrating several
antennas into a laptop computer includes the cost associ-
ated with the antennas themselves, and the modifications to
the computer case, but also the need for additional radio-
frequency (RF) circuitry if the antennas are to work simulta-
neously (in the case of antenna selection, this can be achieved
by switching the antennas on a single RF front-end). The
criteria for the antenna design include performance, inte-
gration, and cost. These parameters were evaluated for each
proposed antenna design. Performance evaluation includes
7
(b)
(b)
the antenna gain, power efficiency and standing wave ratio
(SWR), the aperture of the radiation pattern, and, in the case
of an antenna array, the coupling between antennas. Polar-
ization diversity was also considered, although it was deemed
too costly to exploit. All the antennas developed achieved a
SWR lesser than 2 on the 5.15 to 5.35 GHz band, and a radi-
ation efficiency greater that 90%.
An important effort has also been realized in order to
propose some solutions that could easily be integrated in the
mobile, and that could be produced in large scale in order to
minimize the unit price. Among the considered antenna de-
signs for the MT were monopole and dipole antennas, as well
as a so-called patch antenna, made of folded stamped metal
sheet. The latter, pictured in Figure 7(a), was selected for its
sensitivity, higher than the monopole and dipole alternatives,
and its low space selectivity. Despite these qualities, its manu-
facturing remains simple since it is made out of a single piece
of metal sheet. Moreover, this technology enables the use of
monolithic microwave integrated circuits (MMICs), which
8 EURASIP Journal on Applied Signal Processing
makes putting components on a printed circuit board (PCB)
significantly easier and cheaper in terms of placing and sol-
dering.
For the AP, in order to fit to the different scenarios of
propagation (offices, open space), two types of antennas have
been developed: single-element antennas and antenna ar-
rays. Patch arrays were also considered, for their very high
power efficiency, low manufacturing cost since the technol-
ogy is similar to the PCB fabrication process. Their main
limitation is their poor aperture (lower than 60◦ ). Simple
dipole and dipole arrays (pictured in Figure 7(b)), as well
as dual-polarization dipole arrays were considered. Dipole
arrays overcome the aperture limitation, and, in their dual-
polarization version, can bring in polarization diversity with
very good decoupling (> 20 dB) between polarizations. For
the experimental phase of the project, the single-polarization
dipole array technology was selected.
4. ALGORITHMS
This section describes some of the algorithms considered
for implantation in the MIMO demonstrator. In particu-
lar, we consider space-frequency coding and decoding al-
gorithms and channel estimation and we discuss their re-
spective performance-complexity trade-offs. Due to the wide
scope of this paper and for the sake of space limitation, we
give only a qualitative description of the algorithms. The in-
terested reader will find the appropriate references pointing
to more detailed presentations.
4.1. Coding and decoding
Multiple antennas inherently increase the potential transmis-
sion rate and system diversity. We will then consider coding
and decoding strategies that are able to benefit from the in-
creased number of degrees of freedom in the system. This
section briefly describes the coding and decoding algorithms
studied for implementation in the MIMO demonstrator and
compare them in terms of their performance versus com-
plexity trade-off, with the aid of the specific ray-based chan-
nel simulator for WLAN described in Section 2.1. We first
study open-loop techniques, for which the channel is as-
sumed to be known at the receiver only. Then, we review sev-
eral candidate strategies for the closed-loop case, where the
channel is assumed to be known at both transmitter and re-
ceiver ends. All the algorithms described in this section, use
the underlying convolutional code of rate 1/2 and 64 states
(133, 171)8 used in the HiperLAN/2 standard with packets of
length 384 information bits.
4.1.1. Open loop
We here assume that the channel is perfectly known by the
receiver only, and unknown to the transmitter. We consider
concatenated space-frequency codes with symbol-based pre-
coders, and direct space-frequency codes. In the first class,
the bit stream is first encoded using the standard binary
channel code, and mapped into constellation symbols, then
parsed and possibly further linearly transformed to yield the
actual signal to be transmitted. This class includes the space-
time block coding (STBC) methods developed by Alam-
outi [14], the ABBA method [15], spatial multiplexing as-
sociated to V-BLAST [16], and the linear precoding (LP)
method developed in [17, 18]. On the other hand, direct
space-frequency codes divide the output of the binary code
in subblocks, that are independently interleaved, modulated,
and transmitted over a given (antenna, subcarrier) pair (see,
e.g., [19–21] for more details). This approach is termed
space-time bit-interleaved coded modulation (ST-BICM).
For LP, iterative decoding is performed using minimum
mean-squared error (MMSE) filtering and interference can-
cellation (IC). In the case of ST-BICM, simple interference
cancellation using a matched filter (MF) is used.
4.1.2. Closed loop
Channel state information at the transmitter (CSIT) can pro-
vide substantial benefit in wireless communication systems
compared to the no-CSIT case. CSIT can be obtained ei-
ther through a feedback link, or by using the channel reci-
procity property if both the uplink and downlink operate on
the same frequency in time-division duplex (TDD) systems
as HiperLAN/2.
We next consider coding schemes for multiple-antenna
systems that exploit CSIT. In a nonergodic scenario with per-
fect CSIT, the optimal coding scheme (in the sense of min-
imizing the outage probability) performs the singular value
decomposition (SVD) of the channel, reducing the MIMO
channel to a set of noninterfering parallel channels, and com-
putes the optimal power control rule. Such a coding scheme
yields no-outage, that is, zero error probability at rates below
the so called delayed-limited capacity [22]. However, due to
the large representation size of the channel state and thus,
the high induced complexity, we will focus on suboptimal
schemes only. In particular, we consider the so-called trans-
mit antenna array (TxAA) and selection diversity [23]. In
essence, TxAA transmits over the largest eigenmode of the
channel, while selection diversity transmits over the antenna
whose channel has largest magnitude. As we will see, both
perform very close.
4.1.3. Numerical results
In this section we show some selected numerical exam-
ples using the closed-loop and open-loop techniques de-
scribed above using the ray-tracing simulator described in
Section 2.1. In particular, we show the packet-error rate
(PER) as a function of Eb /N0 = SNR /R per receive antenna,
where R is the spectral efficiency (in bits/s/Hz) and SNR is
the total transmitted signal-to-noise ratio. Iterative methods
employ the BCJR algorithm and five decoding iterations.
Figures 8(a) and 8(b) show the error probability for LP,
V-BLAST, Alamouti, and ST-BICM with NT = NR = 2 an-
tennas and transmission rate R = 1, 2 bit/s/Hz, respectively.
Albert Guillén i Fàbregas et al.
100
10−1
PER
10−2
10−3
10−4
−10 −5 0 5 10 15
Eb /N0 (dB)
LP, BPSK (1 bit/s/Hz)
Alamouti + QPSK (1 bit/s/Hz)
V-BLAST MMSE + BPSK (1 bit/s/Hz)
ST-BICM QPSK SP r = 1/4 MF-IC (1 bit/s/Hz)
(a)
Figure 8: Simulated PER for NT = NR = 2, (a) R = 1 bit/s/Hz and (b) R = 2 bits/s/Hz.
We first observe that ST-BICM performs slightly better
than Alamouti, while the complexity of the latter scheme is
significantly lower. In fact, the complexity of the Alamouti
scheme is dominated by a single run of the Viterbi decoder.
On the other hand, ST-BICM uses iterative interference with
a matched filter, and thus, its complexity is dominated by the
number of times (iterations) the BCJR algorithm is run. We
also see that LP and V-BLAST do not always track the correct
slope, exhibiting nonnegligible losses at PER of interest, for
example, 10−2 . For larger number of antennas, and higher
data rates, ST-BICM provides larger performance advantage.
As stated before, selection diversity and TxAA are sub-
optimal, in the sense that they exploit CSIT to transmit only
over a single mode of the channel, and thus, they implicitly
induce a rate loss of a factor NT . In fact, the weighting opera-
tion can be seen as a linear modification of the channel, creat-
ing a new virtual channel of lower capacity. Figure 9(a) shows
the CDF of the mutual information for the above methods
with NT = NR = 4, perfect CSIT, and SNR = 3 dB. For the
sake of reference, we show the open-loop case (no CSIT) and
two water-filling solutions with short-term power constraint:
global water-filling over the entire dimension of H and a per-
tone water-filling. The global water-filling solution is the op-
timal allocation of power inside an OFDM symbol with in-
stantaneous power constraint [1]. Note that, under short-
term power constraint, CSIT is only relevant at low SNR. Due
to the possibly large dimension of the problem, global water-
filling can be complex. Interestingly, the lower-complexity
per-tone water-filling, where the power is the same in each
frequency subband, and spatial water-filling is used indepen-
dently in each subband, is near-optimal.
9
100
10−1
PER
10−2
10−3
10−4
−10 −5 0 5 10 15 20
Eb /N0 (dB)
LP, QPSK (2 bits/s/Hz)
Alamouti + 16-QAM (2 bits/s/Hz)
V-BLAST MMSE + QPSK (2 bits/s/Hz)
ST-BICM QPSK GR r = 1/2 MF-IC (2 bits/s/Hz)
(b)
Obviously, TxAA and selection diversity are shown to be
largely inferior, due to their rate-one nature. Moreover, we
observe a the degradation incurred by selection diversity, due
to its suboptimality with respect to TxAA.
Figure 9(b) shows the error probability with perfect CSIT
of selection diversity and TxAA with NT = 2 and NR = 1, 2, 4
antennas and transmission rate R = 1 bit/s/Hz. As expected
by the above discussion on outage probability, TxAA outper-
forms selection diversity. However, as selection diversity is
much less complex and more robust to channel estimation
errors, it is a good candidate for practical implementation.
Moreover, as a rate-one scheme, Figure 8(a), selection diver-
sity provides a 1 dB gain with respect to Alamouti and ST-
BICM.
4.2. Synchronization and frequency offset estimation
Synchronization of an MTMR OFDM system is achieved us-
ing training symbols which are simultaneously transmitted
from all transmit antennas at the beginning of each frame.
The HiperLAN/2 broadcast channel (BCH) preamble is de-
scribed in Figure 10. This preamble is divided into 3 fields:
the A-field, for coarse AGC and frame detection, the B-field,
for AGC setting, fine time synchronization, and frequency
offset estimation, and the C-field, which is dedicated to chan-
nel estimation and frequency offset estimation refinement.
HiperLAN/2 usually sets the channel estimation training se-
quences C1 and C2 to be equal. However, as will be shown
next, using different sequences may impact the performance
and complexity of channel estimators for MIMO channels.
Again, any reference to the HiperLAN/2 standard in this
10 EURASIP Journal on Applied Signal Processing

100
1
0.9
0.8 10−1
0.7
0.6

PER
10−2
CDF

0.5

0.4
0.3 10−3

0.2
0.1
10−4
0 −8 −6 −4 −2 0 2 4 6 8 10 12
1 1.5 2 2.5 3 3.5 4 4.5 5 Eb /N0 (dB)
Capacity (bits/s/Hz)
TxAA 2 × 1
No CSIT TxAA 2 × 2
Perfect CSIT, per-tone water filling TxAA 2 × 4
Perfect CSIT, global water filling SD 2 × 1
Perfect CSIT, TxAA SD 2 × 2
Perfect CSIT, selection diversity SD 2 × 4
(a) (b)

Figure 9: (a) Closed loop: mutual information CDF for various closed-loop methods on a 4 × 4 channel at −3 dB SNR and (b) PER of
selection diversity and TxAA for NT = 2 and R = 1 bit/s/Hz.

80 samples 32 samples 128 samples

A-field B-field C32 C64 C64

AGC and coarse AGC setting, frequency Channel estimation

time synchronization offset estimation, and time
synchronization

Figure 10: BCH preamble.

section should be interpreted as a reference to a fairly arbi-
trary OFDM WLAN system for which the values of a num-
ber of parameters are chosen as in the HiperLAN/2 or equiv-
alently IEEE802.11a standards.
A number of different training signal sequences have
been proposed for (1 × 1) OFDM systems such as Hiper-
LAN/2. However, those training symbols, also called pream-
bles, for synchronization and channel estimation, cannot
be applied without care, and potential modifications, to
multiple transmit and receive antennas systems since au-
tocorrelation properties of these sequences are altered by
the contribution of all the channels in the received signal.
Synchronization consists of a four-step process: when the MT
is switched on, it has to
(1) coarsely detect the start of a new frame;
(2) achieve time synchronization;
(3) estimate and compensate frequency offset in the time
domain;
(4) estimate the channel in the frequency domain and per-
form equalization.
Different parts of the OFDM preamble (Figure 10) are used
at each step: the A-field is used for coarse AGC and frame
detection, the B-field is used for AGC setting, fine time syn-
chronization, and frequency offset estimation, and the C-
field for channel estimation and frequency offset estimation
refinement. The synchronization scheme selected for imple-
mentation is based on parallel simultaneous transmission of
the the same A-, B-, and C-fields by all antennas. Therefore,
the autocorrelation properties of the preamble (Figure 11)
remain unchanged at the receiver. We assume that the sig-
nals are received almost simultaneously (with respect to the
sampling frequency), that is, the time instant for received sig-
nals is the same, which is a reasonable assumption given the
small antenna separation. We also exploit the fact that a sin-
gle oscillator is used for both transmit antennas and another
oscillator for both receive antennas, that is, the frequency
Albert Guillén i Fàbregas et al.
1 3.1416
0.8
Magnitude
A-field B-field C-field
0.6
Phase
0.4
0.2
0 −3.1416
96 112 128 144 160 176 192 208 224 240 256 272 288 304 320
Samples
Figure 11: Autocorrelation of BCCH preamble with a 200 kHz frequency offset.
offset is the same for all transmit-receive antenna pairs. In
our demonstrator, the number of unknowns that need to
be estimated is therefore exactly the same as in the single-
antenna case, and we can thus apply the single-antenna syn-
chronization algorithms.
Frequency synchronization consists in finding an esti-
mate of the difference in the frequencies between the trans-
mitter and receiver local oscillators. Frequency offset estima-
tion in OFDM is critical since any frequency offset causes
a loss of subchannel orthogonality, which results in inter-
carrier interference (ICI) and hence performance degrada-
tion. The two main frequency offsets to estimate are carrier
and clock offsets. Due to oscillator inaccuracies, the received
spectrum is usually slightly shifted to the right or left: this
is the so-called carrier frequency offset or frequency offset.
The second error to estimate is due to the drift of the local
oscillator at the receiver: this is the clock frequency offset.
The method considered for carrier frequency offset estima-
tion exploits autocorrelation phase properties. Once the fre-
quency offset has been estimated, it is then compensated in
time domain on the C-field in order to perform channel es-
timation. The selected method is summarized below.
(i) Coarse frequency synchronization: the phase of B-field
peak provides a rough estimate of the frequency offset
within the interval [−625 kHz; 625 kHz[, since the B-
field sequences length is a quarter symbol (autocorre-
lation on 16 samples); there is no ambiguity since the
maximum frequency offset defined in the standard is
200 kHz. However, the degradation of the SNR due to
ICI can still be important.
(ii) Fine frequency synchronization: the C-field sequence
length is one full symbol. If only the C-field was used,
there would be an ambiguity in the frequency offset.
But thanks to the B-field rough estimate, this ambigu-
ity can be resolved, and the SNR degradation becomes
acceptable.
4.3. Channel estimation
In this section, we discuss low-complexity channel esti-
mation algorithms suitable for implantation in the MIMO
testbed. Many pilot-aided channel estimation techniques ex-
ist for single-antenna OFDM-based systems. When directly
exported to the MIMO case, such techniques fail, since the
11
1.5708
0
−1.5708
96 112 128 144 160 176 192 208 224 240 256 272 288 304 320
Samples
signals coming from the different transmit antennas inter-
fere at the receiver end. Moreover, the number of parame-
ters to be estimated can be remarkably larger in the MIMO
case, inducing more complex estimators. More explicitly, in
a single-antenna OFDM system, the number of parameters
to be estimated is NC , while in a multiple-antenna system it
is NC NT NR , though the amount of data also increases by a
factor NR . In this section, we describe how to extend SISO
techniques to the MIMO case while keeping complexity low.
Sending the same pilot symbols simultaneously from all
Tx antennas fails to properly sound the channel, since it only
excites it along a single spatial direction. Therefore, repeat-
ing the same pilot will not bring significant improvement.
On the other hand, joint design of the pilot symbols can be
important. In particular, using orthogonal designs as pilots
[14, 24, 25], and using a training sequence of length L ≥ NT ,
followed by a least-squares estimate in the frequency do-
main, yields an accurate estimate within reasonable complex-
ity. This is possibly the simplest applicable estimation algo-
rithm. Bayesian linear MMSE estimation techniques can also
be used. They rely on the cross-correlation matrix between
the channel and the received signal and the autocorrelation
matrix of the received signal, which in turn depend on the
autocorrelation matrix of the channel elements. These ma-
trices have to be estimated and hence the MMSE methods
are more complex to implement.
In order to reduce the number of parameters to be es-
timated, one can resort to time-domain channel estimation
techniques. In this case, only the coefficients of the channel
impulse response (assumed to be much less than NC ) have to
be estimated. A simple multiplication by the Fourier trans-
form matrix will return the frequency-domain channel pa-
rameters. Unfortunately, however, such techniques yield a
larger overall complexity (see [25] for details). Further de-
tails on these aspects and their exploitation can be found in
[26].
Figure 12 illustrates the performance of several channel
estimators in a MIMO channel with NT = NR = 2 antennas
using the Alamouti scheme, the standard 64 states convolu-
tional code with QPSK modulation (the pilot symbols are
orthogonal and use BPSK) in the BRAN-A channel model
[4]. It can be observed that while the frequency-domain LS
is 2.5 dB away from the perfect channel knowledge case, its
time-domain version (more complex) performs almost as
well as the MMSE.
12 EURASIP Journal on Applied Signal Processing

100

10−1

PER
10−2

10−3
−5 −4 −3 −2 −1 0 1 2 3 4 5
SNR (dB)

Perfect CSIR LS frequency domain

MMSE LS time domain

Figure 12: Performance of different channel estimators with the BRAN-A channel model.

MIMO demonstrator
MIMO transmitter MIMO receiver
PowerquickII-VxWorks real time
PowerquickII-VxWorks real time FPGA MIMO algorithms
development & debug with tornado
development & debug with tornado implementation
Signal ART board ART board Signal
acquisition Remote Remote acquisition
(test vectors) control control (test vectors)

Local Local
oscillators oscillators

Base band
Base band
signal
signal
Antenna 1 Visual C control
Visual C control Antenna 1
software
software Base band Base band
signal Matlab tools
signal RF-FE RF-FE
Antenna 2 Antenna 2
Tx Rx
Remote Remote
control control

C++ simulator
Test vectors Cross verification Test vectors
Performance
Evaluation

Figure 13: Multiple-antenna testbed and performance assessment.

5. MIMO-OFDM DEMONSTRATION TESTBED implemented, in comparison with the single-antenna case.

In this section we describe in some detail the implementation
aspects of the testbed. The main goal of the testbed develop-
ment is the validation of the performance of MIMO algo-
rithmic approaches in real-time hardware, in order to high-
light the potential improvements brought by different modes
We also highlight the methodology used in the testbed de-
velopment, using an FPGA for digital processing. The full
setup is depicted in Figure 13. The steps followed include full
C++ floating-point and fixed-point simulations and cross-
validations with VHDL code simulations. Performance re-
sults have been checked at each step.
Albert Guillén i Fàbregas et al.
After the algorithmic study, a limited set of MIMO al-
gorithms were selected for further implementation, based
on the following criteria: (1) best complexity versus perfor-
mance trade-off, assessed by high-level simulations, (2) ac-
ceptable level of complexity, given the available computa-
tional resources, and (3) embedded reconfigurability with a
single-antenna fallback mode. In the view of the above, the
selected solution for demonstration embeds up to NT = 2 Tx
and NR = 2 Rx antennas. The modulation is based on the
Alamouti space-time block code for the 2 Tx modes, and on
per-subcarrier maximum ratio combining (MRC) for the 2
Rx modes, since it has very low complexity and it still offers
excellent performance on the realistic channel, as shown in
Section 4. The demonstrated system also embeds an OFDM
modulator, with modulation schemes ranging from BPSK to
16-QAM, and forward error correction mechanisms based
on the HiperLAN/2 (or equivalently, IEEE802.11a) specifica-
tions, as well as synchronization and frequency-domain LS
channel estimation procedures. This results in a point-to-
point setup, with embedded performance evaluation mech-
anisms (such as bit error rate computation), usable in differ-
ent locations and scenarios.
5.1. Hardware characteristics
Baseband platform
The algorithms have been ported to the advanced re-
search technologies (ART) platform. The hardware used
in this demonstration platform consists of a radio module
and a baseband processing card. The baseband processing
card consists of a general-purpose microprocessor, for the
non-time-critical MAC code, a large FPGA for the time-
critical data link control (DLC) and physical (PHY) layer
functions, and the analog components needed to interface
to the RF subsystems. Explicitly, there are two digital-to-
analog converters (DACs) and two analog-to-digital convert-
ers (ADCs) on each baseband card. The FPGA is an ALTERA
EP20K1500EBC652-1X (1.5 million gates). It is interfaced
with two 12-bit ADCs and two 12-bit DACs (resp., Ana-
log Devices AD9432BST-105 and AD9752ARU). The FPGA
transmitter and receiver designs include the automatic gain
control (AGC), the synchronization, the OFDM engine, and
the digital I/Q modulator/demodulator. The baseband pro-
cessor is connected to the radio front-end via two AD/DA
converters of 12 bits each, sampled at 80 MHz. The trans-
mitter and receiver designs are based on proprietary blocks
written in VHDL. We choose the development to be per-
formed within three main steps: algorithm-level simulation
(Matlab & C++), then VHDL simulation and FPGA imple-
mentation. The validation of the test bed relies on a cross-
verification: simulation and implementation environments
are compared via common scenarios and bit-accurate test
vectors. The baseband implementation of the receiver is split
into two designs: one is dedicated to the testbed signal ac-
quisition while the other one helps to validate the digital
loop-back process of the cyclic prefix OFDM (CP-OFDM)
13
modulator. In addition, the transmitter uses a frame scenario
player relying on an instruction pipelined sequencer. The sys-
tem can handle BSPK, QPSK, and 16-QAM constellations.
The PCI baseband board is controlled by a C application un-
der MS-Windows. It allows to set any baseband parameters
and frame scenario. On the receiver side, the signal acquisi-
tion and analysis is handled by an additional Matlab envi-
ronment including the OFDM demodulator chain for both,
the CP-OFDM and PRP-OFDM modes. All commands and
transfers initiated by the host PC are relayed by the PowerPC
available on the board; we choose to run this CPU with the
real-time operating system (RTOS) VxWorks. The signal ac-
quisition modules handled by the PowerPC allow to capture
approximately six OFDM frames of 256 data symbols each.
RF front-ends
The architecture of the receiver RF front-end is based on
classical heterodyne (double IF—intermediate frequency)
conversion. It performs the frequency down-conversion only,
since the demodulation is performed digitally in baseband
(digital I/Q separation) on a signal sampled at 80 MHz, with
a lower IF at 20 MHz. At the transmitter, the frequency up-
conversion is achieved with the same principle. The obvious
advantage of this scheme is that only a single baseband pro-
cessing card is needed for the 2 radios. A digital antialias-
ing filter is needed in the receiver to prevent the real and
imaginary parts of the signal from being mixed. If an adja-
cent channel rejection filter is also needed, it must be con-
structed as a band pass filter in at a low IF frequency in
analog. This design is difficult to realize, and thus “digi-
tal I and Q” imposes that the adjacent channel filter must
be performed in the digital domain. Although analog I/Q
would allow to use a lower sampling clock, and therefore re-
duce clock jitter and nonlinearities due to skew, it requires
the compensation of the I/Q imbalances, both in amplitude
and phase, introduced by the separated I/Q processing in
the analog domain. Furthermore, using digital I/Q involves
only one DAC and ADC per radio channel, versus one for
each of the I and Q channels in the analog version. Two lo-
cal oscillators are used for the frequency conversions (the
same for up- and down-conversions) and are based on pro-
grammable PLL synthesizers which enable channel selection
over the HiperLAN/2 band. The demonstrator uses the lower
HiperLAN/2 band (5.180–5.320 GHz), in which the effec-
tive isotropic radiated power (EIRP) is limited to 23 dBm.
The maximum receiving power to be supported according
to the specifications is −30 dBm (class 2 receiver). The mea-
sured performances of the receiver fulfill the sensitivity and
SNR specifications for the various data rates (from 6 Mbit/s
to 54 Mbit/s). The demonstrator, pictured in Figure 14(a),
is built upon purchased components and circuits that were
mounted on dielectric substrates. Some circuits were specif-
ically designed for the RF architecture such as local oscilla-
tors, image rejection filter, and automatic gain control (AGC)
amplifiers at the first and second IF. Compliant with the
IEEE802.11a/HiperLAN/2 definition of the PHY layer, the
14
(a)
Figure 14: RF front-ends. (a) RF-FE stage for 2 Tx/Rx modules. (b) Integrated RF front-end PCB.
demonstrator architecture of the 5 GHz part uses an inter-
mediate fixed frequency at 930 MHz and a voltage-controlled
oscillator (VCO) at 4.2 GHz to address several 20 MHz chan-
nels between 5.15 GHz and 5.35 GHz. On the receiver side,
the chain embeds an AGC with a dynamic range of 80 dB
that pushes the overall sensitivity below −85 dBm. On the
transmitter side, the output power is controlled by an atten-
uator with 15 dB of dynamic range followed by two power
amplifiers (PA) in order to maintain the output power be-
tween −13 dBm and +10 dBm.
In order to reduce cost and more specifically size and vol-
ume, the integration of the RF front-end close to the antenna
is essential. An RF front-end was designed specifically for this
project. Almost all of the RF and IF functions of a single
front-end were implemented on one side of a 10 cm × 15 cm
double-side PCB, with the DC power supply occupying the
other side, together with the synthesizers and the second IF
AGC amplifier. In order to achieve this, new designs of some
circuits were performed: the image rejection filter was de-
signed on a separate high permittivity dielectric substrate
and then mounted on the PCB, the transmitter power am-
plifier, the receiver low-noise amplifier, and the switch were
mounted using separate packaging integrated circuits. The
sheet metal antenna is reported on the opposite side of the RF
circuits. The integrated front-end circuit board is depicted in
Figure 14(b).
Antenna characteristics
The antennas selected for the demonstration platform ad-
dress different sets of requirements, and thus have different
characteristics at the mobile terminal and the access point
(see Figure 7). At the MT, we selected low-cost sheet metal
patch antennas to be integrated into the lid of a laptop,
on the corners, so as to get maximum space diversity (see
Figure 7(a)). They are also oriented so as to get polariza-
tion diversity. The radiation pattern is omnidirectional, and
such antennas present a good radiation efficiency, but a poor
isolation in cross-polarization configuration. It was verified
that in all cases, the coupling between elements is below
EURASIP Journal on Applied Signal Processing
(b)
−30 dB. At the AP, we selected again a low-cost sheet metal
antenna, as baseline single element, having a radiation pat-
tern with 110◦ 3 dB aperture in the vertical plane, and 90◦ in
the horizontal plane, and the two antennas used also present
polarization diversity with 15 dB decoupling. The antenna
gain here is 6 dBi. In addition, we considered a second possi-
bility, which consists in using the baseline elements described
above within an antenna array, having this time a radiation
pattern of (110◦ , 30◦ ) and a 9 dBi gain. Again, it is possible to
use polarization diversity in addition to space diversity, and
the antennas are mounted on a board, so that they can be
tilted (see Figure 7(b)).
5.2. Development methodology
The VHDL design is validated through the following steps:
(1) test VHDL-blocks using software simulation (e.g.,
(i)FFT, Viterbi decoder, synchronization, etc.); all out-
put vectors are compared to bit-accurate test vectors in
C++ and/or Matlab;
(2) synthesize the blocks and check their behavior within
the FPGA on the platform. During this process, the
FPGA outputs must correspond exactly to the results
of the bit-accurate simulations. This approach is valid
for the Tx as well as for the Rx: concerning the Tx,
the signals at the output of the I/Q modulation (which
feed the D/A conversion) are read and validated;
(3) these reference signals are used to test the Rx modules.
They are stored in an internal memory of the platform
and processed by the Rx substituting real received data.
This way, it is ensured that no interference due to D/A
or A/D conversion is present. Processing results of the
Rx baseband blocks are read and compared to the re-
sults of bit-accurate simulations.
The baseband loop-back allows to check the interoperation
of the Tx and Rx without requiring any RF front-end. Never-
theless, a simple connection of the Tx outputs to the Rx in-
puts is insufficient since neither of the analog front-ends deal
Albert Guillén i Fàbregas et al. 15

Table 1: Summary of fixed-point variables precision at the Rx.

Variable Number of bits [bI · bD ]

AGC output [2.9]
Digital I/Q output [3.8]
Antialiasing filter output [3.8]
CIR estimate coefficients [2.6]
FFT input [2.12]
FFT output [2.6]
Demapper distance metric 10

Figure 15: Demonstration scenario for BER measurements.

with I/Q (de)modulation. Thus, digital I/Q (de)modulation
at an IF of 20 MHz is implemented on the baseband platform
using an 80 MHz sampling clock. Hence, the real-valued sig- however, this guard period is not over-dimensioned
nal after the D/A conversion is centered around 20 MHz with for our scenario;
a 20 MHz bandwidth. (iv) the 4 subchannels display very different characteristics.
An extensive evaluation of the precision of the different
The BER was measured with respect to the transmit-
variables involved at the receiver side was performed using
ted power for all supported antenna configurations, and the
a fixed-point C++ simulator. First, each variable is isolated
benefits of the proposed MIMO extensions appear clearly:
by assuming that the rest has full precision. The number of
up to 14.5 dB gain is measured for the 2 Tx, 2 Rx config-
bits used for the fixed-point representation is then adjusted
uration with respect to the reference single-antenna setup.
in order to avoid error propagation and have reasonable per-
These results have been measured for a 16-QAM constella-
formance degradation. We summarize in Table 1 the resolu-
tion and coding rate 1/2, which corresponds to a data rate
tion of the different variables used in the final platform. The
of 24 Mbps, and they are presented in Figure 17. Some com-
values are expressed as [bI · bD ], where bI (resp., bD ) is the
ments on these results are in order. First of all, the perfor-
number of bits for the integer (resp., decimal) part. The over-
mance improvement when going from the 2 × 1 (MISO)
all variable is represented with 1 + bI + bD bits, including the
configuration to the 1 × 2 (SIMO) configuration is 3 dB, as
sign bit.
expected. Second, a large reduction in Tx power required
can be expected at a given BER level, if a large difference
5.3. Overall testbed performance: measurements in in slope is observed. Now, the slopes at the bottom part of
the wireless setup Figure 17 appear to be quite similar for the different cases. It
may be that this impression arises due to the possibility that
The wireless testbed was operated in our lab, within a
the implemented receiver only starts to work well at lower
demonstration scenario aiming at emphasizing the perfor-
BER. This is an effect that would raise the curves at higher
mance improvement brought by multiple antennas in the
BER levels, creating a steeper slope for the lower BER lev-
WLAN context. The exact configuration of the demonstra-
els. Also, as mentioned earlier, each point on these curves has
tion scenario is depicted in Figure 15, with the AP on the left
been obtained with a different channel realization, that prob-
and the MT on the right. Given the presence of lab furniture,
ably did not move much over the 20 frames of data gathered
it was expected that the propagation channel contained mul-
to produce each point. This lack of averaging leads to some
tiple separable paths. Measurements of the MIMO propaga-
variations in the curves. It should be emphasized that the re-
tion channel are plotted in Figure 16, both in frequency do-
sults obtained depend on the positioning of the antennas of
main and time domain. The two curves are the first channel
MT and AP, though the results reported in Figure 17 corre-
response estimated during a given data acquisition period,
spond to a random positioning, that has not at all been cho-
and the last channel response of the same acquisition period,
sen so as to lead to maximum gain. Several other positions
leading to the computation of a given bit error rare (BER)
have been tried, with less extensive measurement results, and
value. Figure 16 highlights several important characteristics
in one other position a gain of 9 dB was observed.
of the propagation at hand in our demonstration context:
(i) during the data acquisition phase leading to the com- 6. CONCLUSIONS
putation of a given BER value, the channel is static;
(ii) the MIMO propagation channel encountered is fre- The goal of the OFDM WLAN testbed developed and dis-
quency selective, with attenuations up to 15 dB from cussed here was to demonstrate the improvement brought by
one carrier to another (power measurements); adding the MIMO dimension to standard WLAN techniques.
(iii) the multipath does not create any additional in- The design of the MIMO testbed was the result of a careful
terblock interference, since it is completely absorbed analysis of and interaction between propagation studies, al-
by the 16-sample (800 nanoseconds) guard interval; gorithm development, and antenna design. The propagation
16 EURASIP Journal on Applied Signal Processing

Tx 1 → Rx 1 Tx 2 → Rx 1
−20 −20

Amplitude (dBV)

Amplitude (dBV)
−30 −30

−40 −40

−50 −50

−60 −60
10 20 30 40 50 10 20 30 40 50
Data carrier number Data carrier number
Tx 1 → Rx 2 Tx 2 → Rx 2
−20 −20
Amplitude (dBV)

Amplitude (dBV)
−30 −30

−40 −40

−50 −50

−60 −60
10 20 30 40 50 10 20 30 40 50
Data carrier number Data carrier number
(a)

Tx 1 → Rx 1 Tx 2 → Rx 1
−30 −30
Amplitude (dBV)

Amplitude (dBV)

−40 −40

−50 −50

−60 −60

−70 −70

−80 −80
10 20 30 40 50 60 10 20 30 40 50 60
Sample number Sample number
Tx 1 → Rx 2 Tx 2 → Rx 2
−30 −30
Amplitude (dBV)
Amplitude (dBV)

−40 −40

−50 −50

−60 −60

−70 −70

−80 −80
10 20 30 40 50 60 10 20 30 40 50 60
Sample number Sample number
(b)

Figure 16: Frequency response and time impulse response of measured realizations of the MIMO propagation channel. (a) Frequency-
domain observation. (b) Time-domain observation.
Albert Guillén i Fàbregas et al.
100
10−1
10−2
BER
10−3
10−4
10−5
−20 −15
Measured transmit power (dBm)
2×2
2×1
Figure 17: BER measured at 24 Mbps for the four antenna configurations versus measured transmit power, 16 QAM symbols.
studies for an indoor environment showed that sufficient
spatial diversity is present (at least at the MT side) to be able
to design algorithms as for the (spatially) i.i.d. Rayleigh case,
for limited MIMO configurations such as 2 × 2 or 4 × 4. The
frequency selectivity on the other hand, even though non-
negligible, turned out to be quite limited, over the frequency
band considered.
These considerations have been taken into account for
the design of innovative space-time coding solutions. To this
end, a wide range of space-time coding algorithms has been
considered and compared. Algorithms can be classified in
terms of the number of streams they transmit (degree of spa-
tial multiplexing) and the degree of channel knowledge that
gets exploited at the transmitter. Another aspect is the degree
of diversity that can be obtained but this is normally max-
imal for the algorithms of interest. Innovative multistream
algorithms have been considered for the case of no chan-
nel knowledge at the transmitter. In particular, ST-BICM has
shown slightly better performance in the 2 × 2 case and sig-
nificantly better performance for larger systems and rates. It
has been observed that the addition of linear precoding leads
to negligible improvement, once channel coding is present.
On the other hand, it was found that channel knowl-
edge at the transmitter leads to little capacity gain at high
SNR. But this high-SNR regime only materializes at higher
SNR as the channel conditioning deteriorates. Single-stream
solutions do not benefit from spatial multiplexing, but the
ensuing loss was found to be very small for the 2 × 2 config-
uration of the testbed. So, for this 2 × 2 case, it was found
that the Alamouti algorithm was close to optimal for the
bit rates considered. When moving to larger MIMO systems,
such as the 4 × 4 case, single-stream solutions become largely
suboptimal. Also multistream solutions that do not bene-
fit from maximal diversity, such as V-BLAST, become quite
17
−10 −5 0 5
1×2
1×1
suboptimal. The ST-BICM approach discussed here then be-
comes very competitive.
The propagation studies also lead to appropriate antenna
designs. At the AP, as the angular aperture is limited, a so-
lution based on a classical antenna array has been proposed
to exploit the spatial diversity. At the MT, where the spatial
diversity is mainly due to angular diversity, pattern diversity
was used. Furthermore, some research was performed into
the optimization of pattern diversity for a larger number of
antennas with practical antenna designs.
The measured results obtained with the 2 × 2 testbed con-
firmed the algorithmic choices made on the basis of analy-
sis and simulations. In particular, the simple Alamouti code
with standard channel coding and simple channel estimation
provides impressive improvements for the 2 × 2 configura-
tion over the 1 × 1 configuration in terms of robustness to the
placement of receive and transmit antennas and general BER.
This improved performance can be exploited to increase bit
rate and/or decrease Tx power and/or increase range. Also,
patch antennas were found to constitute a satisfactory low-
cost solution, allowing for the integration of active RF com-
ponents close to the antennas at the MT.
The developed testbed demonstrates that interesting
MIMO performance can be obtained with readily avail-
able hardware components, reasonable wordlengths, and
overall complexity. Experience with the testbed shows that
the choice of antenna array (pattern and, e.g., polarization
diversity) is a critical issue and that the algorithms may be
usefully adapted to the considered environment.
ACKNOWLEDGMENTS
The research leading to this publication was partially sup-
ported by the French National Telecommunications Research
18
Network (RNRT) project ANTIPODE, by the IST FITNESS
project, and by Antennessa, France Télécom R&D, Eurécom
Institute, Motorola, and ENST Bretagne.
REFERENCES
[1] I. Emre Telatar, “Capacity of multi-antenna Gaussian chan-
nels,” European Transactions on Telecommunications, vol. 10,
no. 6, pp. 585–595, 1999.
[2] G. J. Foschini and M. J. Gans, “On limits of wireless commu-
nications in a fading environment when using multiple an-
tennas,” Wireless Personal Communications, vol. 6, no. 3, pp.
311–335, 1998.
[3] P. Laspougeas, J.-M. Conrat, S. Mayrargue, and P. Pajusco,
“Single and dual multi-sensor channel characterisation in
Indoor and Campus environments at 5 GHz,” in Proceed-
ings of IST Mobile Communications Summit, Barcelona, Spain,
September 2001.
[4] ETSI Normalization Committee, “Channel models for
HIPERLAN/2 in different indoor scenarios,” Norm ETSI
doc. 3ERI085B, European Telecommunications Standards
Institute, Sophia-Antipolis, France, 1998.
[5] P. Laspougeas, P. Pajusco, and J.-C. Bic, “Spatial radio channel
model for UMTS in urban small cells area,” in Proceedings of
European Conference on Wireless Technology (ECWT ’00), pp.
118–121, Paris, France, October 2000.
[6] P. Laspougeas, P. Pajusco, and J.-C. Bic, “Radio propagation
in urban small cells environment at 2 GHz: experimental
spatio-temporal characterization and spatial wideband chan-
nel model,” in Proceedings of 52nd IEEE Vehicular Technology
Conference (VTC ’00), vol. 2, pp. 885–892, Boston, Mass, USA,
September 2000.
[7] J.-M. Conrat and P. Pajusco, “A versatile propagation channel
simulator for MIMO link level simulation,” COST 273 Project
Report TD(03)120, Paris, France, 2003.
[8] V. Erceg, L. Schumacher, P. Kyritsi, et al., “IEEE P802.11
Wireless LANs: TGn Channel Models,” IEEE 802.11 document
03/940r1, 2003.
[9] L. Chaigneaud, V. Guillet, and R. Vauzelle, “A 3D ray-tracing
tool for broadband wireless systems,” in Proceedings of IEEE
VTS 54th Vehicular Technology Conference (VTC ’01), vol. 4,
pp. 2043–2047, Atlantic City, NJ, USA, October 2001.
[10] H. Bölcskei, M. Borgmann, and A. J. Paulraj, “Impact of
the propagation environment on the performance of space-
frequency coded MIMO-OFDM,” IEEE Journal on Selected Ar-
eas in Communications, vol. 21, no. 3, pp. 427–439, 2003, Spe-
cial issue on MIMO systems and applications.
[11] D. Gesbert, H. Bölcskei, D. Gore, and A. J. Paulraj, “MIMO
wireless channels: capacity and performance prediction,” in
Proceedings of IEEE Global Telecommunications Conference
(GLOBECOM ’00), vol. 2, pp. 1083–1088, San Francisco, Calif,
USA, November–December 2000.
[12] D. Chizhik, G. J. Foschini, M. J. Gans, and R. A. Valenzuela,
“Keyholes, correlations, and capacities of multielement trans-
mit and receive antennas,” IEEE Transactions on Wireless Com-
munications, vol. 1, no. 2, pp. 361–368, 2002.
[13] S. Visuri and D. T. M. Slock, “Colocated antenna arrays: de-
sign desiderata for wireless communications,” in Proceedings
of 2nd IEEE Sensor Array and Multichannel Signal Processing
EURASIP Journal on Applied Signal Processing
Workshop (SAM ’02), pp. 580–584, Rosslyn, Va, USA, August
2002.
[14] S. M. Alamouti, “A simple transmit diversity technique for
wireless communications,” IEEE Journal on Selected Areas in
Communications, vol. 16, no. 8, pp. 1451–1458, 1998.
[15] O. Tirkkonen, A. Boariu, and A. Hottinen, “Minimal non-
orthogonality rate 1 space-time block code for 3+ Tx anten-
nas,” in Proceedings of IEEE 6th International Symposium on
Spread Spectrum Techniques and Applications, vol. 2, pp. 429–
432, Parsippany, NJ, USA, September 2000.
[16] P. W. Wolniansky, G. J. Foschini, G. D. Golden, and R. A.
Valenzuela, “V-BLAST: an architecture for realizing very high
data rates over the rich-scattering wireless channel,” in Pro-
ceedings of URSI International Symposium on Signals, Sys-
tems, and Electronics (ISSSE ’98), pp. 295–300, Pisa, Italy,
September–October 1998.
[17] M. Guillaud and D. T. M. Slock, “Full-rate full-diversity space-
frequency coding for MIMO OFDM systems,” in Proceedings
of 3rd IEEE Benelux Signal Processing Symposium (SPS ’02),
Leuven, Belgium, March 2002.
[18] M. Guillaud and D. T. M. Slock, “Multi-stream coding for
MIMO OFDM systems with space-time-frequency spreading,”
in Proceedings of 5th International Symposium on Wireless Per-
sonal Multimedia Communications (WPMC ’02), vol. 1, pp.
120–124, Honolulu, Hawaii, USA, October 2002.
[19] J. Boutros, N. Gresset, and L. Brunel, “Turbo coding & decod-
ing for multiple-antenna channels,” in Proceedings of 3rd In-
ternational Symposium on Turbo Codes and Related Topics, pp.
185–186, Brest, France, September 2003.
[20] A. Guillén i Fàbregas and G. Caire, “Impact of signal con-
stellation expansion on the achievable diversity in quasistatic
multiple-antenna channels,” in Proceedings of 12th European
Signal Processing Conference (EUSIPCO ’04), Vienna, Austria,
September 2004.
[21] A. Guillén i Fàbregas and G. Caire, “Design of space-time
BICM codes for block fading channels with iterative decod-
ing,” in Proceedings of 37th Conference on Information Sciences
and Systems (CISS ’03), Johns Hopkins University Press, Balti-
more, Md, USA, March 2003.
[22] E. Biglieri, G. Caire, and G. Taricco, “Limiting performance of
block-fading channels with multiple antennas,” IEEE Trans-
actions on Information Theory, vol. 47, no. 4, pp. 1273–1289,
2001.
[23] R. Knopp and G. Caire, “Power control and beamforming for
systems with multiple transmit and receive antennas,” IEEE
Transactions on Wireless Communications, vol. 1, no. 4, pp.
638–648, 2002.
[24] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space-time
block codes from orthogonal designs,” IEEE Transactions on
Information Theory, vol. 45, no. 5, pp. 1456–1467, 1999.
[25] A. R. Dias, S. Rouquette, and K. Gosse, “MTMR channel es-
timation and pilot design in the context of space-time block
coded OFDM-based WLANs,” in Proceedings of IST Mobile
and Wireless Telecommunications Summit, pp. 73–77, Thessa-
loniki, Greece, June 2002.
[26] D. T. M. Slock, “Signal processing challenges for wireless com-
munications,” in Proceedings of 1st International Symposium on
Control, Communications and Signal Processing (ISCCSP ’02),
pp. 881–892, Hammamet, Tunisia, March 2004.
Albert Guillén i Fàbregas et al.
Albert Guillén i Fàbregas was born in
Barcelona, Spain, in 1974. He received the
Telecommunications and Electronics Engi-
neering degrees from Universitat Politec-
nica de Catalunya, Barcelona, Spain, and the
Politecnico di Torino, Torino, Italy, respec-
tively, in 1999, and the Ph.D. degree in com-
munication systems from École Polytech-
nique Fédérale de Lausanne (EPFL), Lau-
sanne, Switzerland, in 2004. He conducted
his final research project at the New Jersey Institute of Technol-
ogy (NJIT), Newark, NJ. He has been with the Telecom Italia Re-
search Laboratories, Torino, Italy, and with the European Space
Agency (ESA), Noordwijk, The Netherlands. From 2001 to 2004, he
was a research and teaching assistant at Institut Eurécom, Sophia-
Antipolis, France. From June 2003 to July 2004, he was a visiting
scholar at EPFL. Since September 2004, he has been a Research Fel-
low at the University of South Australia, Mawson Lakes, Australia.
His research areas are communication theory, information theory,
coding theory, and signal processing. He received a Research Fel-
lowship of the Spanish Ministry of Education to join ESA. He re-
ceived the Young Authors Award of the 2004 European Signal Pro-
cessing Conference EUSIPCO 2004, Vienna, Austria, and the 2004
Nokia Best Doctoral Thesis Award from the Spanish Institution of
Telecommunications Engineers.
Maxime Guillaud was born in Bourgoin-
Jallieu, France, in 1977. He received the En-
gineer degree in electronics from École Na-
tionale Supérieure de l’Electronique et de
ses Applications (ENSEA), Cergy, France, in
2000, and the Ph.D. degree in electrical en-
gineering from École Nationale Supérieure
des Télécommunications (ENST), Paris,
France, in 2005. From 2000 to 2001, he was
a Research Engineer at Lucent Technologies’
Bell Laboratories, Holmdel, NJ, USA, where he worked on the pro-
totyping of MIMO UMTS systems. In 2001, he joined the Mo-
bile Communications Department of Eurécom Institute, Sophia-
Antipolis, France. There, he focused on space-time coding and
channel modeling for channels with many degrees of freedom. He
also investigated channel reciprocity and the associated calibration
issues, and introduced the concept of relative calibration. He is the
recipient of a SPAWC ’05 Student Paper Award.
Dirk T. M. Slock received an Engineering
degree from the University of Gent, Bel-
gium, in 1982. In 1984 he was awarded a
Fulbright scholarship from Stanford Uni-
versity, where he received the M.S. de-
gree in EE, M.S. degree in statistics, and
Ph.D. degree in EE in 1986, 1989, and
1989, respectively. While at Stanford, he
developed new fast RLS algorithms for
adaptive filtering. During 1989-91, he was
with Philips Research Belgium. In 1991, he joined the Eu-
recom Institute where he is now Professor. He teaches and
does research on statistical signal processing and its applica-
tion to wireless communications. He invented semiblind equal-
ization, single-antenna interference cancellation for GSM, and the
equalizer-correlator receiver now used in 3G WCDMA terminals.
19
In 2000, he cofounded SigTone, a start-up developing music DSP
product. He has also been a consultant on xDSL and DVB-T sys-
tems. He received one Best Journal Paper Award from the IEEE-
SP and one from EURASIP in 1992. He is the coauthor of two
IEEE Globecom98, one IEEE SIU’04, and one IEEE SPAWC’05 Best
Student Paper Award. He was an Associate editor for the IEEE-SP
Transactions in 1994–96. He is an Editor for EURASIP JASP, for
which he also guest edited two special issues. He is a Fellow of the
IEEE.
Giuseppe Caire was born in Torino, Italy, in
1965. He received the B.S. degree in electri-
cal engineering from Politecnico di Torino
(Italy), in 1990, the M.S. degree in electrical
engineering from Princeton University in
1992, and the Ph.D. degree from Politecnico
di Torino in 1994. He was a recipient of the
AEI G. Someda Scholarship in 1991. He has
been with the European Space Agency (ES-
TEC, Noordwijk, The Netherlands) from
May 1994 to February 1995. Also, he has been an assistant Profes-
sor in telecommunications at the Politecnico di Torino, Associate
Professor at the University of Parma, Italy, Professor at the Insti-
tute Eurecom, Sophia-Antipolis, France, and he is presently Profes-
sor with the Viterbi School of Engineering, University of Southern
California, Los Angeles, Calif He received the Jack Neubauer Best
System Paper Award from the IEEE Vehicular Technology Society
in 2003, the Joint IT/Comsoc Best Paper Award in 2004, and he is a
Fellow of IEEE.
Karine Gosse graduated from the École
Nationale Supérieure des Télécommunica-
tions in 1993, and she got her Ph.D. de-
gree in digital signal processing from the
ENST in 1996. Since then, she has been
working with Motorola Labs in France, on
digital communication systems design, fo-
cussing between 1999 and 2004, as Team
Leader, on multiantenna signal processing
approaches for short-range communication
systems. Her areas of interest include digital signal processing and
digital communications, multicarrier modulation, spatial diversity,
and multiplexing techniques.
Stéphanie Rouquette was born in Figeac,
France, in 1973. She received the Engi-
neering degree from the École Nationale
Supérieure d’Electronique et de Radioélec-
tricité de Bordeaux (ENSERB), Talence,
France, in 1995. She received the Ph.D.
degree from ENSERB in 1998, where she
worked on 2D high-resolution frequency
estimation and radar imaging. In 1999, dur-
ing a postdoctoral year at the Commissariat
à l’Energie Atomique (CEA), Saclay, France, she applied signal pro-
cessing techniques to the functional analysis of magnetic resonance
images. She then joined Motorola Labs, Paris, France, where she is
currently a Senior Research Engineer. Her research interests include
adaptive antennas for wireless communications, OFDM modula-
tion, and channel coding.
20
Alexandre Ribeiro Dias was born in Paris,
France, in 1975. He received an Engi-
neering degree in digital signal process-
ing and system engineering from École
Nationale Supérieure des Télécommunica-
tions, Paris, France, in 2001. Recruited by
Motorola Laboratories, Paris, in 2001, his
research interests were digital signal pro-
cessing for MIMO systems. He was in-
volved in the WLAN standardization activ-
ities of IEEE802.11n, where he made several contributions for
MIMO technologies. He joined Motorola’s Mobile Devices Busi-
ness, Toulouse, France, in 2004, where he is currently a Senior Soft-
ware Engineer working on validation and conformance testing of
GSM/EGPRS cellular phones for new features such as DARP.
Philippe Bernardin graduated from the
École Supérieure d’Ingénieurs en Electron-
ique et Electrotechnique, Marne la Vallée,
France, in 1998, in the specialty of digital
communication systems. After receiving his
degree, he first worked in the Radar Re-
search Department, THALES. He joined the
Motorola Research Center in 2000. Since
then, he has been contributing to wireless
physical layer research studies and particu-
larly worked on real-time hardware implementation. Before 1994,
he was Project Leader in the area of professional phone equip-
ment and particularly he studied and built real-time multispeak-
ers speech recognition systems between 1989 and 1994, in collabo-
ration with the Centre National dÉtude des Télécommunications,
France Telecom. In particular he developed the first commercial
system of speech recognition installed in France for the insurance
company La MACIF. This first activity followed his degree received
from the University of Creteil, France, in 1985.
Xavier Miet received the Engineering de-
gree in electrical engineering and digi-
tal signal processing from École Nationale
Supérieure d’Electronique, Informatique et
Radiocommunications, Bordeaux, France,
in 1999 and the Master of Science de-
gree in digital design and digital commu-
nication from École Nationale Supérieure
des Télécommunications, Paris, France, in
2000. Recruited by Motorola Laboratories
in November 2000, he is currently a Senior Research Engineer. His
recent research interests include VLSI for digital wireless commu-
nication with particular focus on emerging WLAN technologies.
Jean-Marc Conrat was born in Nancy,
France, in 1968. He received the M.S. degree
in electrical engineering from the Applied
Science National Institut (INSA), Lyon,
France, in 1991. Since 1993, he has been
with France Télécom Research and Devel-
opment, Belfort, France. He is involved in
propagation channel measurements, mod-
eling, and simulation. His current research
interests are in the modeling of the direc-
tional wideband propagation channel for MIMO applications.
EURASIP Journal on Applied Signal Processing
Yann Toutain received the Engineer de-
gree from the École Nationale Supérieure
de Télécommunication de Bretagne, Brest,
France, in 1997. Then he received his
Ph.D. degree in electronics from ENSTBr
and Occidental Brittany University, Brest,
France, in 2001. He joined Antennessa,
Brest, France, in 2001 where he spent three
years working on antenna development for
the telecom, automotive, military, and other
specialized markets. He is currently the head of the Dosimetry De-
partment, Antennessa.
Alain Peden received the Diplôme d’Etudes
Supérieures Spécialisées degree from the
University of Limoges, France, in 1985. He
was an MMIC Designer at Thomson Com-
posants Microondes from 1986 to 1988. He
joined the Laboratory of Electronics and
Systems for Telecommunications (LEST),
École Nationale Supérieure des Télécom-
munications in Brest, France, in 1989 and
received the Ph.D. degree in electronics in
1991. He is currently an Associate Professor and his research ac-
tivities are in the field of microwave and millimetre wave power
amplifier design. He is mainly involved with the large-signal char-
acterization of transistors and nonlinear devices and applications
to amplifier design using power combining and quasioptical tech-
niques.
Zaiqing Li was born in Shandong, China,
in 1967. He received the Bachelor’s degree
and Master’s degree in electrical and com-
munication engineering from Harbin Insti-
tute of Technology, in 1989 and 1992, re-
spectively. In 1992, he joined the Depart-
ment of Electrical and Communication En-
gineering, Harbin Institute of Technology,
China. From 2000 to 2001, he was a visit-
ing scholar in the Laboratory of Electron-
ics and Systems for Telecommunications (LEST), Brest, France. He
is currently finishing his Ph.D. degree at the National Engineer
Graduate School (ENST-Bretagne), France. His research interests
involve computational electromagnetics and microwave/millimeter
circuits.
 Photographȱ©ȱTurismeȱdeȱBarcelonaȱ/ȱJ.ȱTrullàs

Preliminaryȱcallȱforȱpapers OrganizingȱCommittee
HonoraryȱChair
The 2011 European Signal Processing Conference (EUSIPCOȬ2011) is the MiguelȱA.ȱLagunasȱ(CTTC)
nineteenth in a series of conferences promoted by the European Association for GeneralȱChair
Signal Processing (EURASIP, www.eurasip.org). This year edition will take place AnaȱI.ȱPérezȬNeiraȱ(UPC)
in Barcelona, capital city of Catalonia (Spain), and will be jointly organized by the GeneralȱViceȬChair
Centre Tecnològic de Telecomunicacions de Catalunya (CTTC) and the CarlesȱAntónȬHaroȱ(CTTC)
Universitat Politècnica de Catalunya (UPC). TechnicalȱProgramȱChair
XavierȱMestreȱ(CTTC)
EUSIPCOȬ2011 will focus on key aspects of signal processing theory and
TechnicalȱProgramȱCo
Technical Program CoȬChairs
Chairs
applications
li ti as listed
li t d below.
b l A
Acceptance
t off submissions
b i i will
ill be
b based
b d on quality,
lit JavierȱHernandoȱ(UPC)
relevance and originality. Accepted papers will be published in the EUSIPCO MontserratȱPardàsȱ(UPC)
proceedings and presented during the conference. Paper submissions, proposals PlenaryȱTalks
for tutorials and proposals for special sessions are invited in, but not limited to, FerranȱMarquésȱ(UPC)
the following areas of interest. YoninaȱEldarȱ(Technion)
SpecialȱSessions
IgnacioȱSantamaríaȱ(Unversidadȱ
Areas of Interest deȱCantabria)
MatsȱBengtssonȱ(KTH)
• Audio and electroȬacoustics.
• Design, implementation, and applications of signal processing systems. Finances
MontserratȱNájarȱ(UPC)
Montserrat Nájar (UPC)
• Multimedia
l d signall processing andd coding.
d
Tutorials
• Image and multidimensional signal processing. DanielȱP.ȱPalomarȱ
• Signal detection and estimation. (HongȱKongȱUST)
• Sensor array and multiȬchannel signal processing. BeatriceȱPesquetȬPopescuȱ(ENST)
• Sensor fusion in networked systems. Publicityȱ
• Signal processing for communications. StephanȱPfletschingerȱ(CTTC)
MònicaȱNavarroȱ(CTTC)
• Medical imaging and image analysis.
Publications
• NonȬstationary, nonȬlinear and nonȬGaussian signal processing. AntonioȱPascualȱ(UPC)
CarlesȱFernándezȱ(CTTC)
Submissions IIndustrialȱLiaisonȱ&ȱExhibits
d i l Li i & E hibi
AngelikiȱAlexiouȱȱ
Procedures to submit a paper and proposals for special sessions and tutorials will (UniversityȱofȱPiraeus)
be detailed at www.eusipco2011.org. Submitted papers must be cameraȬready, no AlbertȱSitjàȱ(CTTC)
more than 5 pages long, and conforming to the standard specified on the InternationalȱLiaison
EUSIPCO 2011 web site. First authors who are registered students can participate JuȱLiuȱ(ShandongȱUniversityȬChina)
in the best student paper competition. JinhongȱYuanȱ(UNSWȬAustralia)
TamasȱSziranyiȱ(SZTAKIȱȬHungary)
RichȱSternȱ(CMUȬUSA)
ImportantȱDeadlines: RicardoȱL.ȱdeȱQueirozȱȱ(UNBȬBrazil)

P
Proposalsȱforȱspecialȱsessionsȱ
l f i l i 15 D 2010
15ȱDecȱ2010
Proposalsȱforȱtutorials 18ȱFeb 2011
Electronicȱsubmissionȱofȱfullȱpapers 21ȱFeb 2011
Notificationȱofȱacceptance 23ȱMay 2011
SubmissionȱofȱcameraȬreadyȱpapers 6ȱJun 2011

Webpage:ȱwww.eusipco2011.org